[Mne_analysis] Maxwell Filter and Projections

Ghuman, Avniel ghumana at upmc.edu
Fri Mar 25 15:13:35 EDT 2016
Search archives: 
And this is why it is pragmatically easier to do it the other way around
(SSP after SSS). Even though the rank of the data is not full after SSS,
SSP does a PCA/SVD so there will be a bunch of eigenvectors with near 0
eigenvalues, which does not cause problems.

In fact, you can use the SVD to estimate the true rank of your data after
SSS, which can be helpful. For example if you want to do multivariate
classification of the MEG data in sensor space, you might be better off
doing it in the eigenspace on only the eigenvectors that do not have near
0 eigenvalues associated with them rather than the full sensor matrix and
the eigenvectors will be orthogonal, so you can use a relatively simpler
classifier that assumes independence, like a Naïve Bayes classifier.

On 3/25/16 3:05 PM, "Matti Hamalainen" <msh at nmr.mgh.harvard.edu> wrote:

>
>In other words, in order for the SSS to work if SSP has been already
>applied, the same SSP projector should be applied to the SSS multipolar
>basis vectors as well.
>
>- Matti
>
>> On Mar 25, 2016, at 3:02 PM, sheraz at nmr.mgh.harvard.edu wrote:
>> 
>> My couple of cents
>> 
>> SSP and SSS (maxfilter) are both spatial filter.
>> 
>> SSS is a single point multipolar expansion using spherical harmonics
>>basis
>> functions,  assumes data to be represented in a particular spatial
>> orientation.
>> 
>> SSP rotates the spatial orientation hence basic assumption of the SSS
>> basis function no longer remain valid.
>> 
>> You can always do SSP after SSS
>> 
>> Sheraz
>> 
>> 
>>> Note, I am not a physicist, but I think this is roughly helpful:
>>> (if a real physicist reads this, please correct me)
>>> 
>>> short version it doesn't make sense, both ssp and sss use the signal
>>> space to remove artifacts
>>> but sss pushes converts the data into spherical harmonics to do this,
>>> but if you reduce the rank
>>> with an ssp vector i don't think going into spherical harmonics would
>>> work.
>>> I think this is also why you have to have a whole head type system to
>>> work in maxfilter in general.
>>> 
>>> hth
>>> d
>>> 
>>> On Fri, Mar 25, 2016 at 2:28 PM, Sam Zorowitz <szorowi1 at gmail.com>
>>>wrote:
>>>> @ Dan: Ahh, we were wondering how the -skip function worked. (We
>>>>looked
>>>> through the manual but did not specify.)
>>>> 
>>>> @ Eric: Absolutely will do.
>>>> 
>>>> All of which brings me back to the original question: Does anyone
>>>>know a
>>>> reason why one shouldn't apply projections before Maxwell filtering?
>>>> 
>>>> On Fri, Mar 25, 2016 at 2:22 PM, Eric Larson <larson.eric.d at gmail.com>
>>>> wrote:
>>>>> 
>>>>> That error is different from the SVD one, but I suspect it has the
>>>>>same
>>>>> underlying cause (Anaconda using outdated build tools). To avoid
>>>>> derailing
>>>>> the original topic too much, can you open an mne-python issue for
>>>>>this
>>>>> specific error scipy.linalg error? We can tackle debugging steps
>>>>>there.
>>>>> If
>>>>> you can anonymize the file, upload it somewhere, give the minimal
>>>>>steps
>>>>> to
>>>>> reproduce it, and paste the output of `mne.sys_info()` (assuming
>>>>>you're
>>>>> on
>>>>> `master`) that would help.
>>>>> 
>>>>> Eric
>>>>> 
>>>>> 
>>>>> On Fri, Mar 25, 2016 at 2:18 PM, Sam Zorowitz <szorowi1 at gmail.com>
>>>>> wrote:
>>>>>> 
>>>>>> @ Dan: We were trying to move away from the command line version
>>>>>>for a
>>>>>> number of reasons. As far as I know, there is no -skip option in the
>>>>>> python
>>>>>> implementation.
>>>>>> 
>>>>>> @ Eric: We are using the Anaconda distribution, and I was not able
>>>>>>to
>>>>>> reproduce the error from the thread you linked. When we are getting
>>>>>>an
>>>>>> error, it looks like this:
>>>>>> 
>>>>>>> RuntimeWarning: invalid value encountered in greater
>>>>>>>  num = np.sum(s > tol, dtype=int)
>>>>>>> 
>>>>>>> 
>>>>>>> 
>>>>>>> Traceback (most recent call last):
>>>>>>>  File "motion_correction.py", line 32, in <module>
>>>>>>>    raw = maxwell_filter(raw, st_duration=st_duration,
>>>>>>> destination=trans.mean(axis=0), bad_condition='warning')
>>>>>>> 
>>>>>>> File "<string>", line 2, in maxwell_filter
>>>>>>> 
>>>>>>> File
>>>>>>> "/homes/5/szoro/.local/lib/python2.7/site-packages/mne/utils.py",
>>>>>>> line 551, in verbose
>>>>>>>    return function(*args, **kwargs)
>>>>>>> 
>>>>>>> File
>>>>>>> 
>>>>>>>"/homes/5/szoro/.local/lib/python2.7/site-packages/mne/preprocessing
>>>>>>>/maxwell.py",
>>>>>>> line 404, in maxwell_filter
>>>>>>>    t_proj = _overlap_projector(orig_in_data, resid, st_correlation)
>>>>>>> 
>>>>>>> File
>>>>>>> 
>>>>>>>"/homes/5/szoro/.local/lib/python2.7/site-packages/mne/preprocessing
>>>>>>>/maxwell.py",
>>>>>>> line 1237, in _overlap_projector
>>>>>>>    overwrite_a=True, mode='economic', **check_disable)[0].T
>>>>>>> 
>>>>>>> File
>>>>>>> 
>>>>>>>"/homes/5/szoro/.local/lib/python2.7/site-packages/scipy/linalg/deco
>>>>>>>mp_qr.py",
>>>>>>> line 142, in qr
>>>>>>>    overwrite_a=overwrite_a)
>>>>>>> 
>>>>>>> File
>>>>>>> 
>>>>>>>"/homes/5/szoro/.local/lib/python2.7/site-packages/scipy/linalg/deco
>>>>>>>mp_qr.py",
>>>>>>> line 20, in safecall
>>>>>>>    ret = f(*args, **kwargs)
>>>>>>> 
>>>>>>> 
>>>>>>> 
>>>>>>> ValueError: failed to create intent(cache|hide)|optional array--
>>>>>>>must
>>>>>>> have defined dimensions but got (0,)
>>>>>> 
>>>>>> 
>>>>>> Unfortunately I can't quite parse what's going on here in the error
>>>>>> message.
>>>>>> 
>>>>>> On Fri, Mar 25, 2016 at 1:47 PM, Eric Larson
>>>>>><larson.eric.d at gmail.com>
>>>>>> wrote:
>>>>>>>> 
>>>>>>>> The short of it is that we were running into some issues with the
>>>>>>>> Maxwell filter function, such that the SVD will not converge for
>>>>>>>> certain
>>>>>>>> buffers if there is an abundance of noise.
>>>>>>> 
>>>>>>> 
>>>>>>> I have observed seen this "failure to converge" problem with SVD
>>>>>>> before.
>>>>>>> Are you running Anaconda? If so, could you look here, see if you
>>>>>>>can
>>>>>>> reproduce the error with the .npy file from the post, and also do
>>>>>>>the
>>>>>>> `cat
>>>>>>> /proc/cpuinfo` step? It will help isolate the problem, and keep
>>>>>>> attention on
>>>>>>> it:
>>>>>>> 
>>>>>>> https://github.com/ContinuumIO/anaconda-issues/issues/695
>>>>>>> 
>>>>>>> If you're not on Anaconda, I'm curious what Python setup you're
>>>>>>> using.
>>>>>>> 
>>>>>>> Such "high noise" segments really shouldn't be a problem for SVD.
>>>>>>>If
>>>>>>> the
>>>>>>> Anaconda folks don't fix the issue, we're going to try to put a
>>>>>>> workaround
>>>>>>> solution in upstream scipy:
>>>>>>> 
>>>>>>> https://github.com/scipy/scipy/pull/5994#issuecomment-200341678
>>>>>>> 
>>>>>>> Eric
>>>>>>> 
>>>>>>> 
>>>>>>> _______________________________________________
>>>>>>> Mne_analysis mailing list
>>>>>>> Mne_analysis at nmr.mgh.harvard.edu
>>>>>>> https://mail.nmr.mgh.harvard.edu/mailman/listinfo/mne_analysis
>>>>>>> 
>>>>>>> 
>>>>>>> The information in this e-mail is intended only for the person to
>>>>>>> whom
>>>>>>> it is
>>>>>>> addressed. If you believe this e-mail was sent to you in error and
>>>>>>> the
>>>>>>> e-mail
>>>>>>> contains patient information, please contact the Partners
>>>>>>>Compliance
>>>>>>> HelpLine at
>>>>>>> http://www.partners.org/complianceline . If the e-mail was sent to
>>>>>>> you
>>>>>>> in error
>>>>>>> but does not contain patient information, please contact the sender
>>>>>>> and
>>>>>>> properly
>>>>>>> dispose of the e-mail.
>>>>>>> 
>>>>>> 
>>>>>> 
>>>>>> _______________________________________________
>>>>>> Mne_analysis mailing list
>>>>>> Mne_analysis at nmr.mgh.harvard.edu
>>>>>> https://mail.nmr.mgh.harvard.edu/mailman/listinfo/mne_analysis
>>>>>> 
>>>>>> 
>>>>>> The information in this e-mail is intended only for the person to
>>>>>>whom
>>>>>> it
>>>>>> is
>>>>>> addressed. If you believe this e-mail was sent to you in error and
>>>>>>the
>>>>>> e-mail
>>>>>> contains patient information, please contact the Partners Compliance
>>>>>> HelpLine at
>>>>>> http://www.partners.org/complianceline . If the e-mail was sent to
>>>>>>you
>>>>>> in
>>>>>> error
>>>>>> but does not contain patient information, please contact the sender
>>>>>> and
>>>>>> properly
>>>>>> dispose of the e-mail.
>>>>>> 
>>>>> 
>>>>> 
>>>>> _______________________________________________
>>>>> Mne_analysis mailing list
>>>>> Mne_analysis at nmr.mgh.harvard.edu
>>>>> https://mail.nmr.mgh.harvard.edu/mailman/listinfo/mne_analysis
>>>>> 
>>>>> 
>>>>> The information in this e-mail is intended only for the person to
>>>>>whom
>>>>> it
>>>>> is
>>>>> addressed. If you believe this e-mail was sent to you in error and
>>>>>the
>>>>> e-mail
>>>>> contains patient information, please contact the Partners Compliance
>>>>> HelpLine at
>>>>> http://www.partners.org/complianceline . If the e-mail was sent to
>>>>>you
>>>>> in
>>>>> error
>>>>> but does not contain patient information, please contact the sender
>>>>>and
>>>>> properly
>>>>> dispose of the e-mail.
>>>>> 
>>>> 
>>>> 
>>>> _______________________________________________
>>>> Mne_analysis mailing list
>>>> Mne_analysis at nmr.mgh.harvard.edu
>>>> https://mail.nmr.mgh.harvard.edu/mailman/listinfo/mne_analysis
>>>> 
>>>> 
>>>> The information in this e-mail is intended only for the person to whom
>>>> it is
>>>> addressed. If you believe this e-mail was sent to you in error and the
>>>> e-mail
>>>> contains patient information, please contact the Partners Compliance
>>>> HelpLine at
>>>> http://www.partners.org/complianceline . If the e-mail was sent to you
>>>> in
>>>> error
>>>> but does not contain patient information, please contact the sender
>>>>and
>>>> properly
>>>> dispose of the e-mail.
>>>> 
>>> _______________________________________________
>>> Mne_analysis mailing list
>>> Mne_analysis at nmr.mgh.harvard.edu
>>> https://mail.nmr.mgh.harvard.edu/mailman/listinfo/mne_analysis
>>> 
>>> 
>>> 
>> 
>> _______________________________________________
>> Mne_analysis mailing list
>> Mne_analysis at nmr.mgh.harvard.edu
>> https://mail.nmr.mgh.harvard.edu/mailman/listinfo/mne_analysis
>> 
>> 
>
>
>_______________________________________________
>Mne_analysis mailing list
>Mne_analysis at nmr.mgh.harvard.edu
>https://mail.nmr.mgh.harvard.edu/mailman/listinfo/mne_analysis

More information about the Mne_analysis mailing list